hattorihanzo1/leonidas-4b-gguf overview

{
  "metadata": {},
  "card_data": {
    "language": [
      "pl",
      "en"
    ],
    "license": "apache-2.0",
    "base_model": "Qwen/Qwen3.5-4B",
    "tags": [
      "llama-cpp",
      "gguf",
      "polish",
      "reasoning",
      "chain-of-thought",
      "lora",
      "mamba",
      "qwen",
      "text-generation",
      "4b"
    ],
    "pipeline_tag": "text-generation",
    "library_name": "gguf",
    "model_type": "qwen3_5",
    "datasets": [
      "michalwlasiuk/data-set-mix"
    ],
    "inference": false,
    "frontmatter": {
      "language": [
        "pl",
        "en"
      ],
      "license": "apache-2.0",
      "base_model": "Qwen/Qwen3.5-4B",
      "tags": [
        "llama-cpp",
        "gguf",
        "polish",
        "reasoning",
        "chain-of-thought",
        "lora",
        "mamba",
        "qwen",
        "text-generation",
        "4b"
      ],
      "pipeline_tag": "text-generation",
      "library_name": "gguf",
      "model_type": "qwen3_5",
      "datasets": [
        "michalwlasiuk/data-set-mix"
      ],
      "inference": "false"
    },
    "hero_image_url": "",
    "summary": "> *\"Come and take them.\"* — Leonidas I, 480 BC **Few parameters. No retreat.** ---",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlanguage:\n  - pl\n  - en\nlicense: apache-2.0\nbase_model: Qwen/Qwen3.5-4B\ntags:\n  - llama-cpp\n  - gguf\n  - polish\n  - reasoning\n  - chain-of-thought\n  - lora\n  - mamba\n  - qwen\n  - text-generation\n  - 4b\npipeline_tag: text-generation\nlibrary_name: gguf\nmodel_type: qwen3_5\ndatasets:\n  - michalwlasiuk/data-set-mix\ninference: false\n---\n\n# Leonidas-4B\n\n> *\"Come and take them.\"* — Leonidas I, 480 BC\n\n**Few parameters. No retreat.**\n\n---\n\n## Model Description\n\nLeonidas-4B is a fine-tuned Polish reasoning model built on the **Qwen3.5-4B hybrid** architecture (Mamba + Attention). It was trained using LoRA fp16 on a curated 48k Polish Chain-of-Thought dataset, with native `<think>` reasoning blocks.\n\nForged by **Hattori Hanzo** — because an idiot admires complexity, a genius admires simplicity.\n\n> *\"An idiot admires complexity, a genius admires simplicity.\"* — Terry A. Davis, TempleOS\n\n---\n\n## Architecture\n\n| Property | Value |\n|---|---|\n| Base Model | Qwen3.5-4B (hybrid Mamba+Attention) |\n| Parameters | ~4B |\n| Training Method | LoRA fp16 (r=16, alpha=16) |\n| Trainable params | 0.47% |\n| Training Steps | 1500 |\n| Final Loss | 0.4389 |\n| Context Length | 32768 |\n| Language | Polish 🇵🇱 + English |\n\n---\n## Training\n\n- **Dataset**: 48k Polish CoT (Chain-of-Thought) — mixed reasoning, math, logic, science\n- **Format**: ChatML with native `<think>` blocks\n- **Platform**: Kaggle T4 16GB\n- **Phases**:\n| Phase | Steps | LR | Scheduler |\n|---|---|---|---|\n| 1 | 2000 | 2e-4 | linear |\n| 2 | 3000 | 5e-5 | cosine |\n| 3 | 5000 | 3e-5 | cosine |\n| 4 | 5000 | 1e-5 | cosine |\n\n---\n## Available Files\n\n| File | Size | Description |\n|---|---|---|\n| `leonidas_f16.gguf` | 8.41 GB | Full precision fp16 |\n| `leonidas_Q8_0.gguf` | 4.47 GB | Q8_0 — best quality |\n| `leonidas_Q6_K.gguf` | 3.45 GB | Q6_K — great quality |\n| `leonidas_Q4_K_M.gguf` | 2.70 GB | Q4_K_M — recommended for most setups |\n\n---\n## Usage\n### llama.cpp\n\n```bash\n./llama-cli \\\n  -m leonidas_Q4_K_M.gguf \\\n  -p \"Jesteś asystentem AI. Myśl krok po kroku.\" \\\n  --chat-template chatml \\\n  -n 1024\n```\n### Ollama / OpenWebUI\nCompatible with any OpenAI-compatible frontend supporting GGUF + ChatML template.\n### Recommended Parameters\n```\ntemperature:        0.6\ntop_p:              0.95\ntop_k:              40\nrepetition_penalty: 1.05\n```\n---\n## Capabilities\n\n- ✅ Native Polish reasoning with `<think>` CoT blocks\n- ✅ Mathematics and logic\n- ✅ Scientific explanations\n- ✅ Sentiment analysis\n- ✅ Code generation (Python, basics)\n- ✅ Quantum physics concepts\n\n---\n## Forge Stamp\n```\ngeneral.author:   HattoriHanzo1\nhanzo.base_model: Qwen3.5-4B\nhanzo.leonidas:   Τότε ἐν τῇ σκιᾷ μαχούμεθα\n                  (\"Then we shall fight in the shade.\")\n```\n---\n## License\nApache 2.0 — base model license from Qwen3.5-4B (Alibaba Cloud).\n---\n\n*Romani ite domum* 😄 — Qapla'! ⚔️🖖",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "llama-cpp",
    "polish",
    "reasoning",
    "chain-of-thought",
    "lora",
    "mamba",
    "qwen",
    "text-generation",
    "4b",
    "pl",
    "en",
    "dataset:michalwlasiuk/data-set-mix",
    "base_model:Qwen/Qwen3.5-4B",
    "base_model:adapter:Qwen/Qwen3.5-4B",
    "license:apache-2.0",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 1496,
  "gated": false,
  "private": false,
  "last_modified": "2026-04-02T17:32:29.000Z",
  "created_at": "2026-03-09T21:55:13.000Z",
  "pipeline_tag": "text-generation",
  "library_name": "gguf"
}

{
  "_id": "69af41c1d286aae3397199f2",
  "id": "HattoriHanzo1/Leonidas-4B-GGUF",
  "modelId": "HattoriHanzo1/Leonidas-4B-GGUF",
  "sha": "904244ad380d58fbb2c347b96d18213916e22cb0",
  "createdAt": "2026-03-09T21:55:13.000Z",
  "lastModified": "2026-04-02T17:32:29.000Z",
  "author": "HattoriHanzo1",
  "downloads": 1496,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "text-generation",
  "library_name": "gguf",
  "siblings_count": 6
}